def test_sparse_2d(self):
p_x = torch.FloatTensor([0.0, 1.0])
H_x = torch.FloatTensor([0.0])
H_x_hat = categorical_entropy(p_x)
self.assertTrue(torch.equal(H_x_hat, H_x))
def test_uniform_2_dists(self):
p_x = torch.FloatTensor([[0.5, 0.5], [0.0, 1.0]])
H_x = torch.FloatTensor([1.0, 0.0])
H_x_hat = categorical_entropy(p_x)
self.assertTrue(torch.equal(H_x_hat, H_x))
def bows_to_ent(bows):
uni_ps = bows_to_ps(bows)
entropies = analysis.categorical_entropy(uni_ps)
avg_entropy = entropies @ bows.sum(dim=1) / bows.sum()
return avg_entropy
def test_nonuniform_3d(self):
p_x = torch.FloatTensor([0.5, 0.25, 0.25])
H_x = torch.FloatTensor([1.5])
self.assertTrue(torch.equal(categorical_entropy(p_x), H_x))